Conceptlanguage-subspace interventions
Language Subspace Intervention
An interpretability and intervention method for multilingual models that suppresses language-related components in intermediate-layer activations through subspace projection, in an attempt to improve cross-lingual reasoning ability.
Timeline (last 90 days)
Nov 4
A new paper examined three key questions across four open-weight large language models using two mathematical reasoning datasets, MATH and MGSM: whether language subspaces capture translation-related variation, whether projection subspaces can remove language information, and whether accuracy can be improved without harming the response language.
Unverified50%